The interactions between caregivers and children play a pivotal role in shaping linguistic competencies of children. One intriguing aspect of this exchange is the feedback provided by caregivers during dialogues, wherein repetition serves as a powerful mechanism for language acquisition. The primary goal of this project is to study patterns of repetition in child-adult dialogues, focusing on the diverse feedback strategies employed by caregivers. These strategies encompass positive repetitions that affirm correct language constructions or corrected versions of the child's attempts.
The primary objectives of this project are:
- Automatic detection of different feedback strategies used by caregivers in child-adult dialogues.
- Deploy LLMs to generate corrective feedback that is both useful and human like.
- Automatic and Human Evaluation to assess correctness and contextual relevance.
- The Talkbank CHILDES Corpora will be used for the project.
- The project will focus on age-specific criteria to ensure relevance to the study of child-adult dialogues.
- The data selection procedure will center around ages 1-4, aligning with the critical developmental period for language acquisition.
- The emphasis will primarily be on positive feedback, this is typically thought of as implicit.
- Positive feedback manifests in the form of modelling examples or corrective feedback.
- To ensure the effectiveness of the approach used, annotations of feedback will be sourced from linguistic experts.
- Large Language Models (LLMs) will be used to generate feedback
- To ensure precision in feedback generation, the project will define a small set of contexts.
- These contexts will serve as the foundation for generating contextually appropriate feedback within the interactions between children and adults.
- To quantitatively evaluate the performance of the feedback generation models, the project will employ the widely recognised automatic metrics: BlEU, BERTscore and MAUVE
- In addition to automated metrics, human evaluators will assess the generated feedback for correctness, appropriateness, and contextual relevance.
- This qualitative assessment ensures a comprehensive understanding of the generated feedback's effectiveness in real-world conversational contexts.
The outcomes of this project are expected to:
- Shed light on how children exploit caregiver feedback to advance their conversational skills.
- Provide insights into the types of repetitions and interactions that can positively influence child language development.
- Have implications for the design of future conversational agents and human-computer interactions.